Cells on Autopilot: Adaptive Cell (Re)Selection via Reinforcement Learning

TL;DR

This paper tackles adaptive optimization of cell (re)selection in heterogeneous 4G/5G networks by formulating the problem as a POMDP and introducing CellPilot, an RL-based framework that outputs Gaussian distributions over six reselection parameters. Through a lightweight MLP policy, seed-specific baselines, curriculum learning, and history stacking, CellPilot achieves substantial gains over a conventional heuristic baseline, including up to $167\%$ throughput improvement and improved load balancing and per-UE efficiency, with strong generalization across unseen areas, loads, and temporal dynamics. Key contributions include problem modeling for hierarchical and equal-priority reselection, a robust training strategy for cross-scenario generalization, and a comprehensive simulation-based evaluation using real-world Brazil network data across multiple geographic scales. The findings demonstrate the practicality of data-driven, autonomous parameter tuning for cell reselection, offering meaningful performance improvements while remaining compatible with existing 4G/5G infrastructure and protocols. This work lays the groundwork for broader RL-driven network optimization, including heterogeneous parameterization and multi-agent coordination, to further enhance mobile network efficiency and user experience.

Abstract

The widespread deployment of 5G networks, together with the coexistence of 4G/LTE networks, provides mobile devices a diverse set of candidate cells to connect to. However, associating mobile devices to cells to maximize overall network performance, a.k.a. cell (re)selection, remains a key challenge for mobile operators. Today, cell (re)selection parameters are typically configured manually based on operator experience and rarely adapted to dynamic network conditions. In this work, we ask: Can an agent automatically learn and adapt cell (re)selection parameters to consistently improve network performance? We present a reinforcement learning (RL)-based framework called CellPilot that adaptively tunes cell (re)selection parameters by learning spatiotemporal patterns of mobile network dynamics. Our study with real-world data demonstrates that even a lightweight RL agent can outperform conventional heuristic reconfigurations by up to 167%, while generalizing effectively across different network scenarios. These results indicate that data-driven approaches can significantly improve cell (re)selection configurations and enhance mobile network performance.

Figures (12)

• Figure 1: Per- downlink throughput comparison across 15 cells in a city in Brazil, under two different parameter configurations for cell (re)selection.
• Figure 2: Large network deployment area used for training and evaluation osm2025.
• Figure 3: Achieved total episode reward over episodes using the baseline area.
• Figure 4: Average values for reward components for the baseline area over different numbers of in the system. The agent is trained on 500 .
• Figure 5: Performance gains for the baseline area and alternative area. The agent is trained on the baseline area and evaluated on both areas to show the cross-area generalizability.